When Is a Draft Accepted? A Theory of Acceptance in Speculative Decoding

arXiv:2606.30265 · cs.LG, cs.CL, stat.ML · Submitted 2026-06-29 · Read on arXiv

Aaryam Sharma

cs.LG, cs.CL, stat.ML

Submitted: 2026-06-29

Comments: 29 pages, 5 figures

Code: https://github.com/NVIDIA/TensorRT-LLM

Project page: https://aaryam.info

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Speculative Decoding — 37× in this paper · explained in 2 episodes
• Monotonicity — 4× in this paper · explained in 2 episodes
• Optimal Transport — 4× in this paper · explained in 5 episodes
• Verification — 4× in this paper · explained in 2 episodes
• Embeddings — 2× in this paper · explained in 2 episodes
• KKT conditions — 2× in this paper · explained in 2 episodes
• Papers — 2× in this paper · explained in 3 episodes
• Transformer — 2× in this paper · explained in 2 episodes
• Verifier — 2× in this paper · explained in 2 episodes
• Absorb — 1× in this paper · explained in 1 episode
• Accuracy — 1× in this paper · explained in 1 episode
• Block Diffusion — 1× in this paper · explained in 2 episodes
• Decoder — 1× in this paper · explained in 2 episodes
• Hyperparameters — 1× in this paper · explained in 2 episodes
• Interpretability — 1× in this paper · explained in 15 episodes
• Sufficiency — 1× in this paper · explained in 2 episodes
• Theoretical Guarantees — 1× in this paper · explained in 2 episodes
• Training Objective — 1× in this paper · explained in 3 episodes
• Transformers — 1× in this paper · explained in 2 episodes

Sources

• DFlash: Block Diffusion for Flash Speculative Decoding
• MARS: Unleashing the Power of Speculative Decoding via Margin-Aware Verification
• Accelerating Production LLMs with Combined Token/Embedding Speculators
• Qwen3 Technical Report
• Speeding up Speculative Decoding via Sequential Approximate Verification

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks